On May 5, 2026, CNN published reporting that Google, Microsoft, and xAI will share unreleased AI model builds with the U.S. government as part of a push to manage cybersecurity-related risk ahead of wide launches. The story attributes the public disclosure to the National Institute of Standards and Technology (NIST).

What CNN reported

  • NIST announced the collaboration; the piece names NIST as the announcing body.
  • The article describes CAISI — the Center for AI Standards and Innovation — as sitting within the U.S. Department of Commerce and being able to evaluate new models and their potential impact on national security and public safety before launch, while also doing follow-on research and testing after deployment.
  • CNN states CAISI has already completed more than 40 AI model evaluations.
  • The story notes broader industry-government context, including Anthropic’s Mythos model coverage and OpenAI describing broader government access for advanced models.

Why it matters

If pre-launch government evaluation becomes a routine checkpoint for frontier-model releases, product and API timelines for major vendors may increasingly interact with standardized third-party testing narratives—not only internal red-teaming or voluntary commitments.

Frequently asked questions

Does this article describe a enacted law?

CNN frames the announcement as agreements and cites NIST and industry participation; readers should distinguish policy reporting from finalized legislation or regulation.

Which exact model builds are covered?

CNN focuses on Google, Microsoft, and xAI sharing unreleased versions; it does not inventory model names in the body of the piece we summarized here.

Where should I read the original wording?

Use CNN’s article text for quotations and fine-grained claims.


Primary source: CNN Business — Microsoft, Google and xAI will let the government test their AI models before launch (May 5, 2026).